[llvm-commits] [Patch] Support for host triple discovery on Windows

Aaron Ballman aaron at aaronballman.com
Fri Oct 14 22:40:35 PDT 2011


On Fri, Oct 14, 2011 at 11:46 PM, NAKAMURA Takumi <geek4civic at gmail.com> wrote:
> Excuse me, I have reverted it in r142055.

I have a revised patch which corrects my misuse of Twine.  I won't
attach it unless we ultimately decide to go with that approach.

> Even if worked, I am dubious for its functionality.
> Is PROCESSOR_LEVEL compatible to gnu style triplet?
> Who affect getHostTriple() and who could give benefits?
>
> Discussion is needed, IMO.

I agree that some discussion is needed.

I tackled the patch due to the FIXME at the top suggesting we derive
the value directly instead of going off the autoconf variable.  But
the real question boils down to: will there be a time when the
autoconf variable doesn't match the execution environment, and
people's expectations?  I don't know what the community expectations
are, so I think we'd need some representative input from others.

I think my biggest worry has to do with MinGW.  If the library was
built with MinGW, then the host triple in config.h comes back as
i686-pc-mingw.  But if I'm running on x86 Windows, I would expect to
see i686-pc-win32 instead.  My secondary, more minor, concern is
whether it would be strange to be running an i786 or i586 and see i686
in the triple.

It could be that my expectations need adjusting though, in which case
perhaps the only thing needed is to remove the FIXME.  What are your
thoughts?

~Aaron



More information about the llvm-commits mailing list